LGAICLFeb 28, 2024

Keeping LLMs Aligned After Fine-tuning: The Crucial Role of Prompt Templates

MILATsinghua
arXiv:2402.18540v2104 citationsh-index: 35NIPS
AI Analysis

This addresses a critical safety issue for users of fine-tuned public LLMs, though it is incremental as it builds on prior findings about alignment loss.

The paper tackles the problem of large language models losing safety alignment after fine-tuning, and finds that using specific prompt templates during fine-tuning and inference can significantly reduce unsafe behaviors, as demonstrated through experiments on models like Llama 2-Chat and GPT-3.5 Turbo.

Public LLMs such as the Llama 2-Chat underwent alignment training and were considered safe. Recently Qi et al. [2024] reported that even benign fine-tuning on seemingly safe datasets can give rise to unsafe behaviors in the models. The current paper is about methods and best practices to mitigate such loss of alignment. We focus on the setting where a public model is fine-tuned before serving users for specific usage, where the model should improve on the downstream task while maintaining alignment. Through extensive experiments on several chat models (Meta's Llama 2-Chat, Mistral AI's Mistral 7B Instruct v0.2, and OpenAI's GPT-3.5 Turbo), this paper uncovers that the prompt templates used during fine-tuning and inference play a crucial role in preserving safety alignment, and proposes the ``Pure Tuning, Safe Testing'' (PTST) strategy -- fine-tune models without a safety prompt, but include it at test time. This seemingly counterintuitive strategy incorporates an intended distribution shift to encourage alignment preservation. Fine-tuning experiments on GSM8K, ChatDoctor, and OpenOrca show that PTST significantly reduces the rise of unsafe behaviors.

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes